Heaven’s Vault is a game built on three pillars - exploration, translation and conversation. The player explores the ancient ruins of the Nebula - they find inscriptions in the script of the Ancients to translate - and then finally, they return to one of the hub moons in the game to share their findings with other characters, in the hope of furthering the game’s main storylines.
The mechanics of the first two pillars are fairly standard. Exploration uses principles of open level design, and translation is essentially puzzle mechanic.
But what about conversation? In a lot of games, conversation is provided as a static menu: a character has topics; the player clicks through them; conversation is complete when all have been clicked. How do we evolve conversation beyond this static paradigm?
Conversation as a Pillar
At inkle, we call the “click through everything” design “a tidying-up mechanic.”
It’s not automatically bad; for some games, it’s the right approach. But it comes with a strong consequence: as soon as the player identifies a tidying-up design, they’ll assume it doesn’t matter in what order they do things, and treat things like tone and conversational style as arbitrary colour. They will expect to achieve complete coverage of the conversation, and expect no risk.
With conversation as a game pillar, we wanted to make dialogue vital, and make characters who are rich, and meaningful, and are full of hidden depths. I outlined the approach we took to Heaven’s Vault’s dialogue in a talk at Adventure X in 2018 entitled “Sparkling Dialogue”, which goes through the process of determining the subtext of each major scene, and then building choices and responses in order to reinforce and develop that subtext. The goal is to create scenes which are dramatically interesting while still giving the player freedom.
However, that strategy only covers half the problem. In interactive scenes, the player is not simply a participant in conversation, they have goals as well - often, they have several. In a game like Heaven’s Vault, the player usually has dozens of “open questions” which they are researching, and will want to be able to ask relevant characters about.
So the deeper writing challenge is this: how do you keep dialogue scene characterful and engaging when the player is actively fighting against your narrative momentum? How do you maintain pace and consequence when what the player wants to do - what they feel they’re supposed to be doing - is going through their notebook of questions and asking every single last one of them?
How do you keep characters feeling juicy, when the player simply wants to squeeze them dry?
Here are some strategies we developed.
1) Know your known knowns and your known unknowns
Heaven’s Vault uses a sophisticated knowledge model to track everything the player has learned across the scope of the game, from minor details like what a character’s name is, through to more esoteric things like “what does the player think happened to the robots after the Fall of the Empire?”
We use this knowledge model to control what questions are available to the player in any given conversation. Questions are only relevant within a window - have you learnt enough to pose the question, and have you not learnt so much that the question is irrelevant?
In particular, this prevents the player asking the same questions over and over: once the answer has been learned from any NPC, then the question becomes unavailable with all NPCs.
Not every game requires a detailed knowledge model but if you’re going to move away from the tidying-up model of investigation, you’re going to need to track something.
2) Say hello!
Dialogue needs structure, even investigative dialogue! Start with an introduction: a greeting, and a first beat of conversation. Give your protagonist an opening question or give your NPC an immediate thing they want to talk about. Have a couple of beats of conversation about something else, before you allow the player to begin asking gameplay-relevant questions.
Player: Good morning, Inspector Lestrade!
Player: And tell me, how is your head this morning?
Lestrade: Better than last night, and no mistake. But I wouldn’t go further than that.
Player: Most regrettable.
Lestrade: Not too regrettable, I hope. But anyway, Holmes, what I can do for you?
This structure has two benefits: firstly, it sets an expectation for the pace of the dialogue - if the player can’t immediately dive into asking their questions, it helps to lay the groundwork for constraints on the dialogue later. Players will be less likely to expect to be able to ask all their questions, or to keep an NPC engaged indefinitely, if a conversation has a more natural beginning.
It also establishes a separate conversational thread that has nothing to do with what the player. Perhaps the NPC is worried that their daughter is late home; perhaps the NPC is suspicious of the player character. Both of those can provide reasons later on for the NPC to push back on questions, to leave early, or to require something else from the player in return for their help.
3) Take Turns!
It’s important to keep the NPC active and alert during a conversation; so they’re not just waiting for the player to give them something to respond to. We build conversations around turn-taking: the player asks a question, the NPC has a response (which might be a one-liner, or might be a sub-conversation); then the NPC gets a turn to say something that’s to their interest.
This helps to keep the conversation moving, but it also provides opportunities for the conversation to go somewhere unexpected and productive. And it’s not dead air: the NPC doesn’t need to be trying to derail the conversation and their contribution doesn’t need to be irrelevant to the wider game! Consider the following dialogue:
Player: What can you tell me about Carstairs?
Lestrade: Not a lot.
Lestrade: But Holmes, I must tell you what we found over at the docklands…
If the dialogue is well-stitched together, the offering from the NPC doesn’t feel unearned even though the player didn’t directly prompt it. It’s simply a reward for engaging in relevant conversation. (The player may not even realise that the NPCs reply wasn’t specific to their question.)
NPCs taking a turn also introduces a nice choice for the player when their turn comes round again. Do they continue the thread introduced by the NPC - which might well be new, and may never appear again - or do they return to their checklist of questions, as they intended? Players will split on a choice like this based on their perception of what’s important in the plot, so it can be a great way to offer some genuine agency.
4) Constrain the Scene
Players who are able to work their way through every possible conversation topic will do so. So pick an arbitrary time limit, and stick to it. Have your NPC walk away, or need to get back to work, or simply get irritated with the protagonist asking so many questions!
This point might feel like the most alien - and perhaps the most obnoxious - but in my opinion, it’s critical. If the player can ask everything, they will stop caring about what they ask next - why does it matter which question they pick, if in the end they’re going to do them all? But if they don’t care about what they’re asking, they won’t care about the answer. The scene will become about clicking through the dialogue as fast as possible: the player will be tidying up with a vacuum cleaner.
But if every conversation ends short - if of your five questions you only ever get three - then every decision about what to ask next is relevant, and every conversation comes with its own in-built tension.
And as with all game design, once you have constraints, you have the opportunity to play with those constraints. If you manage to charm an NPC, perhaps then your limit could be extended. If you irritate them, they might close down more quickly.
Note that this point naturally ties up with the second point above: your introduction can set up the constraint which then curtails the conversation, and the result is quite natural.
5) Constrain the Questions
Sometimes, you’ll only have a few questions that make sense. But what if you have ten or twenty: should we put all twenty in front of the player?
If we do, we offer a high level of agency - but what does it communicate to the player? That they should expect to be able to ask all twenty? Worse, that they’re going to need to work through all twenty? That sounds long and tedious. And then if we constrain the scene and we don’t let the player ask more than a couple of questions, won’t that feel like you missed an awful lot of conversation?
The problem with lots of choices is that they can make the conversation feel automated rather than curated, and players will respond by moving from a narrative mind-set to a systematic one. Players will move from emotionally-resonant choices towards strategic ones, and strip-mine the NPCs replies for salient information instead of treating them as a character’s speech.
But curating a large list of questions sounds like a hard task: how do we know for certain which ones are relevant to the player at the current moment?
In Heaven’s Vault, we use a pretty simple system: a few questions are considered priority topics, and are inserted near the top of the conversation without fail. If you came to talk to Professor Myari about Janniqi Renba, that should be the first question that comes up. After that initial beat, we show any responses that apply to the particular moment in the conversation, shuffle the remaining questions, and limit the final choice to three.
It’s not a perfect solution by any means, and sometimes the balance of what’s high priority and what’s not is hard to strike, but in a game focussed on engaged, narrative conversations, it keeps the momentum of the scene going without sacrificing too much of the player’s ability to follow the threads they care about.
Bringing the Structure Together
Most conversations in Heaven’s Vault follow the macro structure outlined above: introduction, question / NPC remark loop, exit; with the introduction setting up the exit, and with the central loop pushing the player’s knowledge of at least something forwards. The individual elements of conversation then use the ideas outlined in the “Sparkling Dialogue” talk to ensure they feel pacey and characterful.
Some NPCs are more cagey than others, some are down right belligerent, and some will lie, but hopefully all their interactions feel more like authored scenes than drop-down menus.
Investigative conversations ultimately lie on a spectrum: from systematic conversations that tick through lists of questions, to “interactive cutscenes”. Our intention for our “conversation” game pillar was to find a fruitful middle ground: using a back-and-forth between open questions and directed flow to create scenes that are dramatic and engaging while still furthering the game.